MAPSE
A variant of the channel attention module that uses semantic maps from a frozen network to define pooling regions, requiring no additional training of a segmentation network.
Timeline (last 90 days)
The paper proposes four region-aware channel recalibration modules—BASE, SPASE, MAPSE and PLEASE—which define pooling regions using regular grids, SLIC superpixels, frozen network semantic maps and end-to-end learned soft assignments, respectively.
Experimental results show that the four modules surpass the original SE module and related spatial-channel attention variants on most benchmarks, achieving the best results in the majority of cases.
All Facts (2)
The paper proposes four region-aware channel recalibration modules—BASE, SPASE, MAPSE and PLEASE—which define pooling regions using regular grids, SLIC superpixels, frozen network semantic maps and end-to-end learned soft assignments, respectively.
50%UnverifiedExperimental results show that the four modules surpass the original SE module and related spatial-channel attention variants on most benchmarks, achieving the best results in the majority of cases.
50%